It’s largely but not fully deterministic eg backward pass is not deterministic: https://stackoverflow.com/questions/50744565/how-to-handle-non-determinism-when-

It’s largely but not fully deterministic eg backward pass is not deterministic: https://stackoverflow.com/questions/50744565/how-to-handle-non-determinism-when-training-on-a-gpu https://stackoverflow.com/questions/50744565/how-to-handle-n... That said, it’s hard enough to even run an old model 2 years later let alone reproduce to significant digits. I’d be happy enough to solve the former and get “close enough” training results